The paper “Making AI Less ‘Thirsty'” addresses the overlooked water footprint of artificial intelligence (AI), which has been largely neglected amid growing concerns about its carbon footprint. It reveals that training models like GPT-3 can consume significant freshwater resources, with projections indicating AI will require 4.2-6.6 billion cubic meters of water withdrawal by 2027—equivalent to the annual water usage of several countries. This raises alarms given the increasing freshwater scarcity. The authors argue that the AI sector must take social responsibility for its water consumption. They propose a methodology to estimate AI’s water footprint and analyze its unique spatial-temporal water efficiency. By highlighting the need to concurrently manage both water and carbon footprints, the paper advocates for a comprehensive approach to achieving sustainable AI practices in the face of pressing global water challenges.
Source link
